OpenAI Unveils Astra: Next-Gen AI Model Family Solves 10 Open Math and CS Problems

On August 1, 2026, OpenAI officially introduced its next major model family, named Astra. Rather than leading with conventional benchmark scores, OpenAI introduced Astra by demonstrating its capability to solve ten previously open problems in mathematics and theoretical computer science, publishing machine-verifiable Lean proofs directly on GitHub.

A Milestone in Verifiable AI Reasoning

Unlike standard benchmark tests where answers are known ahead of time, Astra solved real open research problems with no prior human solution. Among the achievements was a formal construction proving the existence of non-sofic groups—a long-standing open question in group theory—as well as new upper bounds on sphere-packing density approaching the Cohn-Elkies threshold. The formal proofs were written in Lean, an interactive theorem prover, allowing mathematicians globally to independently verify each step.

Key Highlights of the Astra Announcement

• 10 Verifiable Research Proofs: Solved long-standing open problems across group theory, lattice cryptography, and sphere-packing bounds.
• Formal Lean Verification: Proofs were published as open-source Lean code on GitHub for independent peer verification.
• Cost-Effective Research Scaling: OpenAI estimated the total compute cost to solve all ten problems at approximately $2,000.
• Future Positioning: Astra is positioned as a specialized scientific reasoning engine for complex research workflows.

Implications for AI and Scientific Discovery

The ability to generate machine-verifiable proofs at a compute cost of roughly $2,000 per suite of problems reframes mathematical research as a scalable computational process. Experts note that verifiable proof generation eliminates concerns over hallucinated outputs, establishing a higher bar for evaluating AI reasoning capabilities.

Technical & Model Summary